Intelligent control applications on a binary distillation column

Abstract

Distillation is fundamental in Chemical Engineering. It is a highly complex and no-linear process. Therefore, developing intelligent control systems for distillation columns is challenging. These control techniques are based on previous knowledge and intuitive rules. In this work, several intelligent control strategies, such as Expert, Fuzzy (Mamdani and Sugeno) and Neural-Network Control are applied to control a simulated distillation column, and their performance compared with a traditional PI controller. The controlled variable was the distillate molar fraction and the manipulated variable was the reflux ratio. All control strategies were tested in two scenarios: i) set-point changes, and ii) disturbance and setpoint changes. The best control strategy was the Neural-Network, using a NARMA-L2 controller. This control has a good disturbance rejection and a fast set-point tracking with a smooth control action.